SGDMA has corrupted read data in 8-bit operation

[Altera_Forum]

I'm using the sgdma with 32-bit aligned read buffers as recommended but I still get corrupted read data. I haven't found any known issues about 8-bit operation. Does anyone know if 8-bit operation works? It seems to work for many transfers but not all as if the starting address affects it. 

 Thanks

[Altera_Forum]

--- Quote Start ---  I'm using the sgdma with 32-bit aligned read buffers as recommended but I still get corrupted read data. I haven't found any known issues about 8-bit operation. Does anyone know if 8-bit operation works? It seems to work for many transfers but not all as if the starting address affects it.  Thanks  --- Quote End ---    What interface are you transferring the data over?  I had terrible problems with holes in DMA'd data (from my dma engine) across the PCIe bus in Q 11.X with a memory mapped avalon interface.  Q12.0 fixed the problem (although the Q12.0 qsys component editor's badly broken so I'd recommend Q12.1 ).   Nial

[Altera_Forum]

If the data is coming from, or being verified by, a nios cpu, are you sure the data cache isn't affecting things?

[Altera_Forum]

GZoinker - Thanks for the suggestion. I was monitoring the read data directly (via SignalTap) going into the DMA so I know the data is bad from the start. 

 dsl - I think you may have got the answer! I'll check it out. Thanks!

[Altera_Forum]

Everyone gets caught out by the lack of cache coherency! 

(except those of us who only use tightly coupled memory and no data caches)

[Altera_Forum]

Thanks dsl, clearing the cache has cleared things up! 

 Cheers Greg